Ino Load Lora Clip Model

Most LoRA loaders ask you to pick from a dropdown of files already in models/loras. Ino Load Lora Clip Model instead takes a path string for the LoRA - which is exactly what you want when the file doesn't exist yet. This node pairs with the Ino pack's download chain: InoCivitaiDownloadModel (or a HuggingFace download node) writes a LoRA to disk and hands you its abs_path, you feed that path straight into this node's model_path, and it loads what just landed. It's a building block for self-bootstrapping workflows where the model and its LoRA are fetched on the fly.

It applies the LoRA to both model and CLIP - hence the name - which is the standard behavior for character/style LoRAs that carry their own text-encoder adjustments.

How it works

It takes model_path, strips the extension and uses the filename as the LoRA name, then calls ComfyUI's core LoraLoader with that path and your model + clip. The extra output you don't get from the stock node: a trigger_word. The author derives it by splitting the filename on the first underscore (filename_part1_part2 → trigger filename_part1). That's a heuristic, not metadata - it assumes you name your LoRAs triggername_rest-of-name.safetensors, which is common convention but not universal. If the trigger looks wrong, check the filename, because that's all the node has to go on.

It also returns success and message booleans/strings so downstream logic can react to a failed load instead of silently passing a None through - the classic ComfyUI footgun where a missing LoRA produces a broken graph with no error until the sampler dies.

Inputs and outputs that matter

• model, clip - wire from your checkpoint loaders.
• model_path - the string path to the .safetensors. Absolute path works; a filename-relative path generally needs to resolve inside your loras folder, which is where the download nodes put things anyway.
• strength_model / strength_clip - both default 1.0, fine-stepped. Same dual-strength logic as any modern LoRA loader.

Outputs: success, message, name (the derived LoRA name), trigger_word, and the patched model + clip for your sampler.

Installing it

ComfyUI Manager → search "ComfyUI Ino Nodes" → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/nobandegani/comfyui_ino_nodes
cd comfyui_ino_nodes
pip install -r requirements.txt

Restart after. Requires inopyutils (handled by requirements.txt) and a current ComfyUI on the V3-style node API. No API keys needed for loading - only if you're using it with the download nodes that need a Civitai/HF token.

Common issues

Most failures trace back to the path. If you're typing a path by hand, use an absolute path or a path that resolves inside ComfyUI/models/loras; the download nodes return a working abs_path so the safest route is to wire that output in rather than hand-typing. And remember the trigger word is derived from the filename - if you renamed a LoRA, the trigger it reports changed with it, even though the actual trigger baked into the LoRA's training data didn't. The success output exists precisely so you don't have to debug a silent None: check it, and fail loudly.